Final Project

Preparing R Studio

Install Libraries

library(vosonSML)
library(tidyverse)
library(tidytext)
library(quanteda)
library(syuzhet)
library(reshape2)
library(lda)
library(jsonlite)
library(stringr)
library(tm)
library(rtweet)
library(ggplot2)
library(lubridate)
library(hms)
library(caTools)
library(caret)
library(e1071)
library(rpart)
library(rpart.plot)
library(party)
library(forecast)
library(arules)

AWS

RDBS Connection

host = "text-analysis.cpkypoj1gkct.us-east-1.rds.amazonaws.com"
port = 3306
user = "rvillarreal"
password = "x10809877"

my_instance = DBI :: dbConnect(
  RMySQL :: MySQL(),
  host = host,
  port = port,
  user = user,
  password = password
)

Creating Database

dbSendQuery(my_instance, "CREATE DATABASE TWEETS")

Database Connection

dbname = "TWEETS"

con = DBI :: dbConnect(
  RMySQL :: MySQL(),
  dbname = dbname,
  host = host,
  port = port,
  user = user,
  password = password
)

Twitter

API Connection

consumer_key <- "R7gjfyZMLN2vK6iFlQf2hhCPw"
consumer_secret <- "Zr5BqyPKPzkfMpAi3HijHk5HNFgNffvisPDAiSwd9fLyDKB3Il"
access_token <- "1326312317705916416-6estSlWdrKq26zO5blGoD1OaOVfLCV"
access_secret <- "Jh4UgN83EZWV4wLQmjkJZkKdZz0ByNe5kl3q1qHowb81L"

twitter_token <- create_token(
  app = "Programming_Data_Analytics",
  consumer_key = consumer_key,
  consumer_secret = consumer_secret,
  access_token = access_token,
  access_secret = access_secret
)

setup_twitter_oauth(consumer_key, consumer_secret, access_token, access_secret)

[1] "Using direct authentication"

Tweets Extraction and CSV

#messi_tweets9 <- search_tweets("messi OR MESSI OR Messi", n = 18000, lang = "en", include_rts = F)

messi_tweets_final <- data.frame(unique(rbind(messi_tweets3, messi_tweets2, messi_tweets, messi_tweets4, messi_tweets5, messi_tweets6, messi_tweets7, messi_tweets8, messi_tweets9)))

messi_tweets_final <- messi_tweets_final %>%
                      separate(created_at, into = c("Date", "Time"), sep = " ", convert = TRUE)

messi_tweets_final$Time <- as.hms(messi_tweets_final$Time)
messi_tweets_final$Date <- as.Date(messi_tweets_final$Date)

write.csv(apply(messi_tweets_final,2,as.character), "/Users/renevillarreal/Desktop/STU/Programming for Data Analytics/messi_tweets_final.csv")

Tweets Final Data Frame

Messi_Final_Project <- read.csv("messi_tweets_final.csv")
Messi_Final_Project

Final Dataset for Analysis

Storing & Extracting from Database

#dbCreateTable(con, "fct_tweets", Messi_Final_Project, row.names = NULL)
#dbWriteTable(con, "fct_tweets", Messi_Final_Project, append = T, row.names = F)

#final_data <- dbReadTable(con, "fct_tweets")
#final_data

Prepare Dataset

final_data <- final_data %>% 
              separate(created_at, into = c("Date", "Time"), sep = " ", convert = TRUE)

Expected 2 pieces. Missing pieces filled with `NA` in 78237 rows [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, ...].

final_data$Time <- as.hms(final_data$Time)
final_data$Date <- as.Date(final_data$Date)

final_data <- final_data %>% 
              select(user_id, status_id, Date, Time, screen_name, text, source, display_text_width, reply_to_status_id, is_quote, 
                     favorite_count, retweet_count, hashtags, symbols, lang, name, location, description, protected, followers_count, 
                     friends_count, listed_count, statuses_count, favourites_count, account_created_at, verified, reply_to_status_id)

final_data

Analysis

Problem Statement

Classify if a tweet was made from an Apple device or not given the follower count and friends count. Determine which model is best using Accuracy, Precision, and Recall.

Logistic Regression Model

# Data Prep
final_data_logistic <- mutate(final_data, Source_Class = ifelse(source == "Twitter for iPhone" | source == "Twitter for iPad" | source == "Twitter for Mac" | source == "Twitter for i?S", 
                                                "Apple", "Non_Apple")
                     )

final_data_logistic$Source_Class <- factor(final_data_logistic$Source_Class, levels=c("Non_Apple", "Apple"))


# Sample, Test & Train
set.seed(305)

final_data_logistic$sample <- sample.split(final_data_logistic$Source_Class, SplitRatio = .80)

train_log <- subset(final_data_logistic, sample == TRUE)
test_log <- subset(final_data_logistic, sample == FALSE)


# GLM Model
log_model <- glm(formula = Source_Class ~ followers_count + friends_count,
               data = train_log,
               family = binomial)

glm.fit: fitted probabilities numerically 0 or 1 occurred

# Prediction
test_log$SourceProbability <- predict(log_model, test_log, type = "response")


# Classification
test_log <- mutate(test_log,
               PredictedSource = ifelse(SourceProbability < 0.5, "Non_Apple", "Apple"))

test_log$PredictedSource <- factor(test_log$PredictedSource, levels = c("Non_Apple", "Apple"))


# Metrics
Accuracy_Log <- confusionMatrix(test_log$Source_Class, test_log$PredictedSource)[["overall"]][["Accuracy"]]
Precision_Log <- precision(test_log$Source_Class, test_log$PredictedSource)
Recall_Log <- recall(test_log$Source_Class, test_log$PredictedSource)

cat("\n", "The Accurracy of the Logistic Regression Model is:", Accuracy_Log, "\n")

 The Accurracy of the Logistic Regression Model is: 0.6316227 

cat("The Precision of the Logistic Regression Model is:", Precision_Log, "\n")

The Precision of the Logistic Regression Model is: 1 

cat("The Recall of the Logistic Regression Model is:", Recall_Log, "\n")

The Recall of the Logistic Regression Model is: 0.6316227 

CTREE Model

# Data Prep
final_data_ctree <- mutate(final_data, Source_Class = ifelse(source == "Twitter for iPhone" | source == "Twitter for iPad" | source == "Twitter for Mac" | source == "Twitter for i?S", 
                                                "Apple", "Non_Apple")
                     )

final_data_ctree$Source_Class <- factor(final_data_ctree$Source_Class, levels=c("Non_Apple", "Apple"))


# Sample, Test & Train
set.seed(305)

final_data_ctree$sample <- sample.split(final_data_ctree$Source_Class, SplitRatio = .80)

train_ctree <- subset(final_data_ctree, sample == TRUE)
test_ctree <- subset(final_data_ctree, sample == FALSE)


# CTREE Model
ctree_model <- ctree(formula = Source_Class ~ followers_count + friends_count, train_ctree)


# Plot Model
plot(ctree_model)

# Prediction
Predicted_CTREE <- predict(ctree_model, test_ctree)


# Metrics
Accuracy_CTREE <- confusionMatrix(test_ctree$Source_Class, Predicted_CTREE)[["overall"]][["Accuracy"]]
Precision_CTREE <- precision(test_ctree$Source_Class, Predicted_CTREE)
Recall_CTREE <- recall(test_ctree$Source_Class, Predicted_CTREE)

cat("The Accurracy of the CTREE Model is:", Accuracy_CTREE, "\n")

The Accurracy of the CTREE Model is: 0.6644085 

cat("The Precision of the CTREE Model is:", Precision_CTREE, "\n")

The Precision of the CTREE Model is: 0.9480927 

cat("The Recall of the CTREE Model is:", Recall_CTREE, "\n")

The Recall of the CTREE Model is: 0.6641622 

CART Model

# Data Prep
final_data_cart <- mutate(final_data, Source_Class = ifelse(source == "Twitter for iPhone" | source == "Twitter for iPad" | source == "Twitter for Mac" | source == "Twitter for i?S", 
                                                "Apple", "Non_Apple")
                     )

final_data_cart$Source_Class <- factor(final_data_cart$Source_Class, levels=c("Non_Apple", "Apple"))


# Sample, Test & Train
set.seed(305)

final_data_cart$sample <- sample.split(final_data_cart$Source_Class, SplitRatio = .80)

train_cart <- subset(final_data_cart, sample == TRUE)
test_cart <- subset(final_data_cart, sample == FALSE)


# CART Model
cart_model <- rpart(formula = Source_Class ~ followers_count + friends_count, train_cart)


# Plot Model
rpart.plot(cart_model)

# Prediction
Predicted_CART <- predict(cart_model, test_cart, type = "class")
table(test_cart$Source_Class, Predicted_CART, dnn = c("Actual", "Prediction"))

           Prediction
Actual      Non_Apple Apple
  Non_Apple      9883     0
  Apple          5764     0

# Metrics
Accuracy_CART <- confusionMatrix(test_cart$Source_Class, Predicted_CART)[["overall"]][["Accuracy"]]
Precision_CART <- precision(test_cart$Source_Class, Predicted_CART)
Recall_CART <- recall(test_cart$Source_Class, Predicted_CART)

cat("The Accurracy of the CART Model is:", Accuracy_CART, "\n")

The Accurracy of the CART Model is: 0.6316227 

cat("The Precision of the CART Model is:", Precision_CART, "\n")

The Precision of the CART Model is: 1 

cat("The Recall of the CART Model is:", Recall_CART, "\n")

The Recall of the CART Model is: 0.6316227 

Naive Bayes Model

# Data Prep
final_data_nb <- mutate(final_data, Source_Class = ifelse(source == "Twitter for iPhone" | source == "Twitter for iPad" | source == "Twitter for Mac" | source == "Twitter for i?S", 
                                                "Apple", "Non_Apple")
                     )

final_data_nb$Source_Class <- factor(final_data_nb$Source_Class, levels=c("Non_Apple", "Apple"))


# Sample, Test & Train
set.seed(305)

final_data_nb$sample <- sample.split(final_data_nb$Source_Class, SplitRatio = .80)

train_nb <- subset(final_data_nb, sample == TRUE)
test_nb <- subset(final_data_nb, sample == FALSE)


# Naive Bayes Model
nb_model <- naiveBayes(Source_Class ~ followers_count + friends_count, train_nb)


# Prediction
Predicted_NB <- predict(nb_model, test_nb, type = "class")
table(test_nb$Source_Class, Predicted_NB, dnn = c("Actual", "Prediction"))

           Prediction
Actual      Non_Apple Apple
  Non_Apple       357  9526
  Apple           142  5622

# Metrics
Accuracy_NB <- confusionMatrix(test_nb$Source_Class, Predicted_NB)[["overall"]][["Accuracy"]]
Precision_NB <- precision(test_nb$Source_Class, Predicted_NB)
Recall_NB <- recall(test_nb$Source_Class, Predicted_NB)

cat("The Accurracy of the Naive Bayes Model is:", Accuracy_NB, "\n")

The Accurracy of the Naive Bayes Model is: 0.382118 

cat("The Precision of the Naive Bayes Model is:", Precision_NB, "\n")

The Precision of the Naive Bayes Model is: 0.03612263 

cat("The Recall of the Naive Bayes Model is:", Recall_NB, "\n")

The Recall of the Naive Bayes Model is: 0.7154309 

Conclusion

The CART Model is the best model to use. It has the highest Accuracy and Recall, with the second highest Precision.

Time Series

# Time Series Data Frame
final_data_ts <- messi_tweets_final %>%
                 mutate(source2 = ifelse(source == "Twitter for iPhone" | source == "Twitter for iPad" | source == "Twitter for Mac" | source == "Twitter for i?S", "Apple", "Non_Apple")) %>%
                 filter(source2 == "Apple") %>%
                 select(source, source2, Date, Time) %>%
                 arrange(Date, Time)


# Hour and Date Breakout
final_data_ts$Time <- hour(as.hms(final_data_ts$Time))
final_data_ts$Date <- as.numeric(as.Date(final_data_ts$Date))


# Counting Instances
final_data_ts <- final_data_ts %>%
  arrange(Date, Time) %>%
  group_by(Date, Time) %>%
  summarize(count= n())

`summarise()` regrouping output by 'Date' (override with `.groups` argument)

# Time Series Model
model_ts <-  ts(final_data_ts$count, start = c(18588, 16), end = c(18602, 23), frequency = 24)


# Decomposition
plot(decompose(model_ts))

# ARIMA Model
arima_messi <- auto.arima(model_ts)


# Prediction: 7 days
forecast_ts <- forecast(arima_messi, h = 168)


# Forecast Plot
plot(forecast_ts)

NA
NA
NA